Dimitris Vatousis

Hardware, software, and the space between.

Dimitris Vatousis works across electronics, embedded systems, machine learning and production software. He currently builds roaming platforms for EV charging at Parity Platform (EV Loader) in Athens.

Get in touch View projects
Portrait of Dimitris Vatousis
Athens, Greece

Experience

Parity Platform
EV Loader · Athens, Greece · Hybrid
  1. Software DeveloperFull-time · Aug 2026 to present
  2. Software DeveloperInternship · Jun 2026 to Aug 2026

I work on the company's OCPI 2.2.1 roaming platform, which connects EV charge points to the mobility providers whose drivers use them.

  • Built the eMSP role from scratch, designed to run alongside the existing CPO platform and work with it from day one.
  • Found and fixed critical defects in the CPO platform affecting sessions, commands, token authorization and partner synchronisation.
  • Built the web operations consoles for both roles: live charging sessions, partner traffic tracing, finance and revenue reporting, and audited staff actions.
  • Built the payment terminal module, which lets drivers pay for charging directly at the charge point, including invoicing.
  • Hardened the platform through repeated security and compliance audits.

Projects

Showing all 6 projects

EV Loader OCPI roaming platform

Production platform for EV charging roaming at Parity Platform: the eMSP role, operations consoles for both roles, and payment terminals.

OCPI 2.2.1eMSPCPOPayments
Professional work, private codePlatforms

Energy efficient LLM inference on embedded hardware

Experiment software that measures real power draw while running small language models with quantization, pruning and dynamic precision scaling.

Pythonllama-cpp-pythonPyTorchRaspberry Pi 400
View codeML · Embedded

Kaggle: Child Mind Institute, Problematic Internet Use

Bronze medal solution, placing 180th out of 3,560 teams.

PythonKaggle
View codeML

SatCore: satellite mission control and telemetry

Open source ground segment toolkit: spacecraft health monitoring, telemetry forecasting, pass scheduling from real TLE data for 14,000+ satellites, and live 3D orbits.

PythonStreamlitPlotlyMIT
View codeSpace

Line follower robot

Designed, built and programmed a line following robot with Arduino and infrared sensors for a university robotics competition.

ArduinoC++IR sensors
View codeEmbedded

ARM NN and PyArmNN on Raspberry Pi

An up to date guide (July 2025) to building ARM NN with TensorFlow Lite and PyArmNN support, from dependencies to inference with a real TFLite model.

ARM NNTensorFlow LiteRaspberry Pi
Read the guideEmbedded · ML

Research

  1. 2026
    Undergraduate dissertation
    Democritus University of Thrace
    Thesis

    Research and Development of Energy-Efficient Inference Methods for Generative AI (LLMs) on Low-Power Embedded Systems

    A comparison of ways to run large language models on low power hardware, measuring what each one really costs in energy with an external hardware power meter instead of software estimates.

    Hardware
    Raspberry Pi 400, Atorch UD18 USB power meter
    Models
    TinyLlama 1.1B, Llama 3.2 1B, Qwen2 0.5B
    Methods
    fp16, INT8 (Q8_0), 30% unstructured pruning, dynamic precision scaling
    Supervisor
    Assoc. Prof. Dimitrios Karampatzakis
    • INT8 quantization cut energy per generated token by about 45% across all three models.
    • Power draw stayed nearly constant at 8.1 to 8.5 W, so energy is set mainly by how long inference runs.
    • Model size affected energy more than any optimisation technique. Pruning without retraining gave no benefit.
    Read the thesisOriginal, Greek · PDF Read the thesisEnglish translation · PDF
    Experiment code

About

Dimitris Vatousis works across hardware and software: electronics, embedded systems and robotics on one side, machine learning and production platforms on the other.

He studied in the Department of Informatics at the Democritus University of Thrace (DUTH) in Greece, and works at Parity Platform (EV Loader) in Athens.

His interests include electronics, embedded systems, robotics, machine learning and network security.

Get in touch

Open to conversations about work, research and collaboration.